conversation-summarization
Pass
Audited by Gen Agent Trust Hub on Apr 3, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The
ai_summarizecode template inSKILL.mdpresents an indirect prompt injection surface where external conversation transcripts are processed without adequate safeguards. - Ingestion points: The
conversation.full_transcriptvariable is interpolated directly into an LLM prompt inSKILL.md. - Boundary markers: The prompt template lacks delimiters or boundary markers to separate user-provided transcript data from system instructions.
- Capability inventory: The skill describes integrating these summaries with CRM systems (Salesforce, HubSpot) and automated notifications, which could be exploited if the transcript contains malicious instructions.
- Sanitization: No sanitization or validation of the transcript content is implemented or recommended in the provided example.
Audit Metadata